pix2pix-tensorflow

This repository is a Tensorflow implementation of the Isola's Image-to-Image Tranaslation with Conditional Adversarial Networks, CVPR2017.

Requirements

• tensorflow 1.8.0
• python 3.5.3
• numpy 1.14.2
• matplotlib 2.0.2
• scipy 0.19.0

Generated Results

• facades dataset
  A to B: from RGB image to generate label image
  B to A: from label image to generate RGB image

• maps dataset A to B: from statellite image to generate map image B to A: from map image to generate statellite image

Generator & Discriminator Structure

• Generator structure

• Discriminator structure

Documentation

Download Dataset

Download datasets (script borrowed from torch code)

bash ./src/download_dataset.sh [dataset_name]

• dataset_name supports cityscapes, edges2handbags, edges2shoes, facades, and maps.
  Note: our implementation has tested on facades and maps dataset only. But you can easily revise the code to run on other datasets.

Directory Hierarchy

├── pix2pix
│   ├── src
│   │   ├── dataset.py
│   │   ├── download_dataset.sh
│   │   ├── main.py
│   │   ├── pix2pix.py
│   │   ├── solver.py
│   │   ├── tensorflow_utils.py
│   │   └── utils.py
├── Data
│   ├── facades
│   └── maps

Note: please put datasets on the correct position based on the Directory Hierarchy.

Training pix2pix Model

Use main.py to train a pix2pix model. Example usage:

python main.py --dataset=facades --which_direction=0 --is_train=true

• gpu_index: gpu index, default: 0
• dataset: dataset name for choice [facades|maps], default: facades
• which_direction: AtoB (0) or BtoA (1), default: AtoB 0
• batch_size: batch size for one feed forward, default: 1
• is_train: 'training or inference mode, default: False
• learning_rate: initial learning rate, default: 0.0002
• beta1: momentum term of Adam, default: 0.5
• iters: number of interations, default: 200000
• print_freq: print frequency for loss, default: 100
• save_freq: save frequency for model, default: 20000
• sample_freq: sample frequency for saving image, default: 500
• sample_batch: sample size for check generated image quality, default: 4
• load_model: folder of save model that you wish to test, (e.g. 20180704-1736). default: None

Evaluating pix2pix Model

Use main.py to evaluate a pix2pix model. Example usage:

python main.py --is_train=false --load_model=folder/you/wish/to/test/e.g./20180704-1746

Please refer to the above arguments.

Citation

  @misc{chengbinjin2018pix2pix,
    author = {Cheng-Bin Jin},
    title = {pix2pix tensorflow},
    year = {2018},
    howpublished = {\url{https://github.com/ChengBinJin/pix2pix-tensorflow}},
    note = {commit xxxxxxx}
  }

Attributions/Thanks

• This project borrowed some code from yenchenlin and pix2pix official websit
• Some readme formatting was borrowed from Logan Engstrom

License

Copyright (c) 2018 Cheng-Bin Jin. Contact me for commercial use (or rather any use that is not academic research) (email: sbkim0407@gmail.com). Free for research use, as long as proper attribution is given and this copyright notice is retained.

Related Projects

• DiscoGAN